iterative-retrieval

To provide relevant codebase context to subagents, use an iterative retrieval loop that progressively discovers and refines files through relevance scoring and dynamic query refinement across up to three cycles.

Handling missing-context failures in multi-agent workflows requires a four-phase loop: Dispatch, Evaluate, Refine, and Loop, which identifies context gaps and dynamically refines subsequent searches to resolve missing information.

Iterative retrieval reduces context size for large codebases by applying relevance scoring and gap identification after each dispatch, ensuring only high-relevance files are passed forward to subagents.

Yes, iterative retrieval supports codebase exploration for bug fixes, feature implementations, and codebase audits where relevant file patterns and context unfold progressively across multiple search cycles.

The iterative retrieval pattern integrates file-pattern search directly into its multi-agent context management workflow, using dynamic query refinement to locate and score relevant files for subagent dispatch.

The iterative retrieval loop limits context discovery to a maximum of three cycles, meaning highly complex or deeply nested codebases may require manual intervention if relevant context is not identified within those iterations.